]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/vdev_indirect_mapping.c
Linux 4.18 compat: inode timespec -> timespec64
[mirror_zfs.git] / module / zfs / vdev_indirect_mapping.c
1 /*
2 * CDDL HEADER START
3 *
4 * This file and its contents are supplied under the terms of the
5 * Common Development and Distribution License ("CDDL"), version 1.0.
6 * You may only use this file in accordance with the terms of version
7 * 1.0 of the CDDL.
8 *
9 * A full copy of the text of the CDDL should have accompanied this
10 * source. A copy of the CDDL is also available via the Internet at
11 * http://www.illumos.org/license/CDDL.
12 *
13 * CDDL HEADER END
14 */
15
16 /*
17 * Copyright (c) 2015 by Delphix. All rights reserved.
18 */
19
20 #include <sys/dmu_tx.h>
21 #include <sys/dsl_pool.h>
22 #include <sys/spa.h>
23 #include <sys/vdev_impl.h>
24 #include <sys/vdev_indirect_mapping.h>
25 #include <sys/zfeature.h>
26 #include <sys/dmu_objset.h>
27
28 #ifdef ZFS_DEBUG
29 static boolean_t
30 vdev_indirect_mapping_verify(vdev_indirect_mapping_t *vim)
31 {
32 ASSERT(vim != NULL);
33
34 ASSERT(vim->vim_object != 0);
35 ASSERT(vim->vim_objset != NULL);
36 ASSERT(vim->vim_phys != NULL);
37 ASSERT(vim->vim_dbuf != NULL);
38
39 EQUIV(vim->vim_phys->vimp_num_entries > 0,
40 vim->vim_entries != NULL);
41 if (vim->vim_phys->vimp_num_entries > 0) {
42 ASSERTV(vdev_indirect_mapping_entry_phys_t *last_entry =
43 &vim->vim_entries[vim->vim_phys->vimp_num_entries - 1]);
44 ASSERTV(uint64_t offset =
45 DVA_MAPPING_GET_SRC_OFFSET(last_entry));
46 ASSERTV(uint64_t size = DVA_GET_ASIZE(&last_entry->vimep_dst));
47
48 ASSERT3U(vim->vim_phys->vimp_max_offset, >=, offset + size);
49 }
50 if (vim->vim_havecounts) {
51 ASSERT(vim->vim_phys->vimp_counts_object != 0);
52 }
53
54 return (B_TRUE);
55 }
56 #endif
57
58 uint64_t
59 vdev_indirect_mapping_num_entries(vdev_indirect_mapping_t *vim)
60 {
61 ASSERT(vdev_indirect_mapping_verify(vim));
62
63 return (vim->vim_phys->vimp_num_entries);
64 }
65
66 uint64_t
67 vdev_indirect_mapping_max_offset(vdev_indirect_mapping_t *vim)
68 {
69 ASSERT(vdev_indirect_mapping_verify(vim));
70
71 return (vim->vim_phys->vimp_max_offset);
72 }
73
74 uint64_t
75 vdev_indirect_mapping_object(vdev_indirect_mapping_t *vim)
76 {
77 ASSERT(vdev_indirect_mapping_verify(vim));
78
79 return (vim->vim_object);
80 }
81
82 uint64_t
83 vdev_indirect_mapping_bytes_mapped(vdev_indirect_mapping_t *vim)
84 {
85 ASSERT(vdev_indirect_mapping_verify(vim));
86
87 return (vim->vim_phys->vimp_bytes_mapped);
88 }
89
90 /*
91 * The length (in bytes) of the mapping object array in memory and
92 * (logically) on disk.
93 *
94 * Note that unlike most of our accessor functions,
95 * we don't assert that the struct is consistent; therefore it can be
96 * called while there may be concurrent changes, if we don't care about
97 * the value being immediately stale (e.g. from spa_removal_get_stats()).
98 */
99 uint64_t
100 vdev_indirect_mapping_size(vdev_indirect_mapping_t *vim)
101 {
102 return (vim->vim_phys->vimp_num_entries * sizeof (*vim->vim_entries));
103 }
104
105 /*
106 * Compare an offset with an indirect mapping entry; there are three
107 * possible scenarios:
108 *
109 * 1. The offset is "less than" the mapping entry; meaning the
110 * offset is less than the source offset of the mapping entry. In
111 * this case, there is no overlap between the offset and the
112 * mapping entry and -1 will be returned.
113 *
114 * 2. The offset is "greater than" the mapping entry; meaning the
115 * offset is greater than the mapping entry's source offset plus
116 * the entry's size. In this case, there is no overlap between
117 * the offset and the mapping entry and 1 will be returned.
118 *
119 * NOTE: If the offset is actually equal to the entry's offset
120 * plus size, this is considered to be "greater" than the entry,
121 * and this case applies (i.e. 1 will be returned). Thus, the
122 * entry's "range" can be considered to be inclusive at its
123 * start, but exclusive at its end: e.g. [src, src + size).
124 *
125 * 3. The last case to consider is if the offset actually falls
126 * within the mapping entry's range. If this is the case, the
127 * offset is considered to be "equal to" the mapping entry and
128 * 0 will be returned.
129 *
130 * NOTE: If the offset is equal to the entry's source offset,
131 * this case applies and 0 will be returned. If the offset is
132 * equal to the entry's source plus its size, this case does
133 * *not* apply (see "NOTE" above for scenario 2), and 1 will be
134 * returned.
135 */
136 static int
137 dva_mapping_overlap_compare(const void *v_key, const void *v_array_elem)
138 {
139 const uint64_t * const key = v_key;
140 const vdev_indirect_mapping_entry_phys_t * const array_elem =
141 v_array_elem;
142 uint64_t src_offset = DVA_MAPPING_GET_SRC_OFFSET(array_elem);
143
144 if (*key < src_offset) {
145 return (-1);
146 } else if (*key < src_offset + DVA_GET_ASIZE(&array_elem->vimep_dst)) {
147 return (0);
148 } else {
149 return (1);
150 }
151 }
152
153 /*
154 * Returns the mapping entry for the given offset.
155 *
156 * It's possible that the given offset will not be in the mapping table
157 * (i.e. no mapping entries contain this offset), in which case, the
158 * return value value depends on the "next_if_missing" parameter.
159 *
160 * If the offset is not found in the table and "next_if_missing" is
161 * B_FALSE, then NULL will always be returned. The behavior is intended
162 * to allow consumers to get the entry corresponding to the offset
163 * parameter, iff the offset overlaps with an entry in the table.
164 *
165 * If the offset is not found in the table and "next_if_missing" is
166 * B_TRUE, then the entry nearest to the given offset will be returned,
167 * such that the entry's source offset is greater than the offset
168 * passed in (i.e. the "next" mapping entry in the table is returned, if
169 * the offset is missing from the table). If there are no entries whose
170 * source offset is greater than the passed in offset, NULL is returned.
171 */
172 static vdev_indirect_mapping_entry_phys_t *
173 vdev_indirect_mapping_entry_for_offset_impl(vdev_indirect_mapping_t *vim,
174 uint64_t offset, boolean_t next_if_missing)
175 {
176 ASSERT(vdev_indirect_mapping_verify(vim));
177 ASSERT(vim->vim_phys->vimp_num_entries > 0);
178
179 vdev_indirect_mapping_entry_phys_t *entry = NULL;
180
181 uint64_t last = vim->vim_phys->vimp_num_entries - 1;
182 uint64_t base = 0;
183
184 /*
185 * We don't define these inside of the while loop because we use
186 * their value in the case that offset isn't in the mapping.
187 */
188 uint64_t mid;
189 int result;
190
191 while (last >= base) {
192 mid = base + ((last - base) >> 1);
193
194 result = dva_mapping_overlap_compare(&offset,
195 &vim->vim_entries[mid]);
196
197 if (result == 0) {
198 entry = &vim->vim_entries[mid];
199 break;
200 } else if (result < 0) {
201 last = mid - 1;
202 } else {
203 base = mid + 1;
204 }
205 }
206
207 if (entry == NULL && next_if_missing) {
208 ASSERT3U(base, ==, last + 1);
209 ASSERT(mid == base || mid == last);
210 ASSERT3S(result, !=, 0);
211
212 /*
213 * The offset we're looking for isn't actually contained
214 * in the mapping table, thus we need to return the
215 * closest mapping entry that is greater than the
216 * offset. We reuse the result of the last comparison,
217 * comparing the mapping entry at index "mid" and the
218 * offset. The offset is guaranteed to lie between
219 * indices one less than "mid", and one greater than
220 * "mid"; we just need to determine if offset is greater
221 * than, or less than the mapping entry contained at
222 * index "mid".
223 */
224
225 uint64_t index;
226 if (result < 0)
227 index = mid;
228 else
229 index = mid + 1;
230
231 ASSERT3U(index, <=, vim->vim_phys->vimp_num_entries);
232
233 if (index == vim->vim_phys->vimp_num_entries) {
234 /*
235 * If "index" is past the end of the entries
236 * array, then not only is the offset not in the
237 * mapping table, but it's actually greater than
238 * all entries in the table. In this case, we
239 * can't return a mapping entry greater than the
240 * offset (since none exist), so we return NULL.
241 */
242
243 ASSERT3S(dva_mapping_overlap_compare(&offset,
244 &vim->vim_entries[index - 1]), >, 0);
245
246 return (NULL);
247 } else {
248 /*
249 * Just to be safe, we verify the offset falls
250 * in between the mapping entries at index and
251 * one less than index. Since we know the offset
252 * doesn't overlap an entry, and we're supposed
253 * to return the entry just greater than the
254 * offset, both of the following tests must be
255 * true.
256 */
257 ASSERT3S(dva_mapping_overlap_compare(&offset,
258 &vim->vim_entries[index]), <, 0);
259 IMPLY(index >= 1, dva_mapping_overlap_compare(&offset,
260 &vim->vim_entries[index - 1]) > 0);
261
262 return (&vim->vim_entries[index]);
263 }
264 } else {
265 return (entry);
266 }
267 }
268
269 vdev_indirect_mapping_entry_phys_t *
270 vdev_indirect_mapping_entry_for_offset(vdev_indirect_mapping_t *vim,
271 uint64_t offset)
272 {
273 return (vdev_indirect_mapping_entry_for_offset_impl(vim, offset,
274 B_FALSE));
275 }
276
277 vdev_indirect_mapping_entry_phys_t *
278 vdev_indirect_mapping_entry_for_offset_or_next(vdev_indirect_mapping_t *vim,
279 uint64_t offset)
280 {
281 return (vdev_indirect_mapping_entry_for_offset_impl(vim, offset,
282 B_TRUE));
283 }
284
285
286 void
287 vdev_indirect_mapping_close(vdev_indirect_mapping_t *vim)
288 {
289 ASSERT(vdev_indirect_mapping_verify(vim));
290
291 if (vim->vim_phys->vimp_num_entries > 0) {
292 uint64_t map_size = vdev_indirect_mapping_size(vim);
293 vmem_free(vim->vim_entries, map_size);
294 vim->vim_entries = NULL;
295 }
296
297 dmu_buf_rele(vim->vim_dbuf, vim);
298
299 vim->vim_objset = NULL;
300 vim->vim_object = 0;
301 vim->vim_dbuf = NULL;
302 vim->vim_phys = NULL;
303
304 kmem_free(vim, sizeof (*vim));
305 }
306
307 uint64_t
308 vdev_indirect_mapping_alloc(objset_t *os, dmu_tx_t *tx)
309 {
310 uint64_t object;
311 ASSERT(dmu_tx_is_syncing(tx));
312 uint64_t bonus_size = VDEV_INDIRECT_MAPPING_SIZE_V0;
313
314 if (spa_feature_is_enabled(os->os_spa, SPA_FEATURE_OBSOLETE_COUNTS)) {
315 bonus_size = sizeof (vdev_indirect_mapping_phys_t);
316 }
317
318 object = dmu_object_alloc(os,
319 DMU_OTN_UINT64_METADATA, SPA_OLD_MAXBLOCKSIZE,
320 DMU_OTN_UINT64_METADATA, bonus_size,
321 tx);
322
323 if (spa_feature_is_enabled(os->os_spa, SPA_FEATURE_OBSOLETE_COUNTS)) {
324 dmu_buf_t *dbuf;
325 vdev_indirect_mapping_phys_t *vimp;
326
327 VERIFY0(dmu_bonus_hold(os, object, FTAG, &dbuf));
328 dmu_buf_will_dirty(dbuf, tx);
329 vimp = dbuf->db_data;
330 vimp->vimp_counts_object = dmu_object_alloc(os,
331 DMU_OTN_UINT32_METADATA, SPA_OLD_MAXBLOCKSIZE,
332 DMU_OT_NONE, 0, tx);
333 spa_feature_incr(os->os_spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
334 dmu_buf_rele(dbuf, FTAG);
335 }
336
337 return (object);
338 }
339
340
341 vdev_indirect_mapping_t *
342 vdev_indirect_mapping_open(objset_t *os, uint64_t mapping_object)
343 {
344 vdev_indirect_mapping_t *vim = kmem_zalloc(sizeof (*vim), KM_SLEEP);
345 dmu_object_info_t doi;
346 VERIFY0(dmu_object_info(os, mapping_object, &doi));
347
348 vim->vim_objset = os;
349 vim->vim_object = mapping_object;
350
351 VERIFY0(dmu_bonus_hold(os, vim->vim_object, vim,
352 &vim->vim_dbuf));
353 vim->vim_phys = vim->vim_dbuf->db_data;
354
355 vim->vim_havecounts =
356 (doi.doi_bonus_size > VDEV_INDIRECT_MAPPING_SIZE_V0);
357
358 if (vim->vim_phys->vimp_num_entries > 0) {
359 uint64_t map_size = vdev_indirect_mapping_size(vim);
360 vim->vim_entries = vmem_alloc(map_size, KM_SLEEP);
361 VERIFY0(dmu_read(os, vim->vim_object, 0, map_size,
362 vim->vim_entries, DMU_READ_PREFETCH));
363 }
364
365 ASSERT(vdev_indirect_mapping_verify(vim));
366
367 return (vim);
368 }
369
370 void
371 vdev_indirect_mapping_free(objset_t *os, uint64_t object, dmu_tx_t *tx)
372 {
373 vdev_indirect_mapping_t *vim = vdev_indirect_mapping_open(os, object);
374 if (vim->vim_havecounts) {
375 VERIFY0(dmu_object_free(os, vim->vim_phys->vimp_counts_object,
376 tx));
377 spa_feature_decr(os->os_spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
378 }
379 vdev_indirect_mapping_close(vim);
380
381 VERIFY0(dmu_object_free(os, object, tx));
382 }
383
384 /*
385 * Append the list of vdev_indirect_mapping_entry_t's to the on-disk
386 * mapping object. Also remove the entries from the list and free them.
387 * This also implicitly extends the max_offset of the mapping (to the end
388 * of the last entry).
389 */
390 void
391 vdev_indirect_mapping_add_entries(vdev_indirect_mapping_t *vim,
392 list_t *list, dmu_tx_t *tx)
393 {
394 vdev_indirect_mapping_entry_phys_t *mapbuf;
395 uint64_t old_size;
396 uint32_t *countbuf = NULL;
397 vdev_indirect_mapping_entry_phys_t *old_entries;
398 uint64_t old_count;
399 uint64_t entries_written = 0;
400
401 ASSERT(vdev_indirect_mapping_verify(vim));
402 ASSERT(dmu_tx_is_syncing(tx));
403 ASSERT(dsl_pool_sync_context(dmu_tx_pool(tx)));
404 ASSERT(!list_is_empty(list));
405
406 old_size = vdev_indirect_mapping_size(vim);
407 old_entries = vim->vim_entries;
408 old_count = vim->vim_phys->vimp_num_entries;
409
410 dmu_buf_will_dirty(vim->vim_dbuf, tx);
411
412 mapbuf = vmem_alloc(SPA_OLD_MAXBLOCKSIZE, KM_SLEEP);
413 if (vim->vim_havecounts) {
414 countbuf = vmem_alloc(SPA_OLD_MAXBLOCKSIZE, KM_SLEEP);
415 ASSERT(spa_feature_is_active(vim->vim_objset->os_spa,
416 SPA_FEATURE_OBSOLETE_COUNTS));
417 }
418 while (!list_is_empty(list)) {
419 uint64_t i;
420 /*
421 * Write entries from the list to the
422 * vdev_im_object in batches of size SPA_OLD_MAXBLOCKSIZE.
423 */
424 for (i = 0; i < SPA_OLD_MAXBLOCKSIZE / sizeof (*mapbuf); i++) {
425 vdev_indirect_mapping_entry_t *entry =
426 list_remove_head(list);
427 if (entry == NULL)
428 break;
429
430 uint64_t size =
431 DVA_GET_ASIZE(&entry->vime_mapping.vimep_dst);
432 uint64_t src_offset =
433 DVA_MAPPING_GET_SRC_OFFSET(&entry->vime_mapping);
434
435 /*
436 * We shouldn't be adding an entry which is fully
437 * obsolete.
438 */
439 ASSERT3U(entry->vime_obsolete_count, <, size);
440 IMPLY(entry->vime_obsolete_count != 0,
441 vim->vim_havecounts);
442
443 mapbuf[i] = entry->vime_mapping;
444 if (vim->vim_havecounts)
445 countbuf[i] = entry->vime_obsolete_count;
446
447 vim->vim_phys->vimp_bytes_mapped += size;
448 ASSERT3U(src_offset, >=,
449 vim->vim_phys->vimp_max_offset);
450 vim->vim_phys->vimp_max_offset = src_offset + size;
451
452 entries_written++;
453
454 vmem_free(entry, sizeof (*entry));
455 }
456 dmu_write(vim->vim_objset, vim->vim_object,
457 vim->vim_phys->vimp_num_entries * sizeof (*mapbuf),
458 i * sizeof (*mapbuf),
459 mapbuf, tx);
460 if (vim->vim_havecounts) {
461 dmu_write(vim->vim_objset,
462 vim->vim_phys->vimp_counts_object,
463 vim->vim_phys->vimp_num_entries *
464 sizeof (*countbuf),
465 i * sizeof (*countbuf), countbuf, tx);
466 }
467 vim->vim_phys->vimp_num_entries += i;
468 }
469 vmem_free(mapbuf, SPA_OLD_MAXBLOCKSIZE);
470 if (vim->vim_havecounts)
471 vmem_free(countbuf, SPA_OLD_MAXBLOCKSIZE);
472
473 /*
474 * Update the entry array to reflect the new entries. First, copy
475 * over any old entries then read back the new entries we just wrote.
476 */
477 uint64_t new_size = vdev_indirect_mapping_size(vim);
478 ASSERT3U(new_size, >, old_size);
479 ASSERT3U(new_size - old_size, ==,
480 entries_written * sizeof (vdev_indirect_mapping_entry_phys_t));
481 vim->vim_entries = vmem_alloc(new_size, KM_SLEEP);
482 if (old_size > 0) {
483 bcopy(old_entries, vim->vim_entries, old_size);
484 vmem_free(old_entries, old_size);
485 }
486 VERIFY0(dmu_read(vim->vim_objset, vim->vim_object, old_size,
487 new_size - old_size, &vim->vim_entries[old_count],
488 DMU_READ_PREFETCH));
489
490 zfs_dbgmsg("txg %llu: wrote %llu entries to "
491 "indirect mapping obj %llu; max offset=0x%llx",
492 (u_longlong_t)dmu_tx_get_txg(tx),
493 (u_longlong_t)entries_written,
494 (u_longlong_t)vim->vim_object,
495 (u_longlong_t)vim->vim_phys->vimp_max_offset);
496 }
497
498 /*
499 * Increment the relevant counts for the specified offset and length.
500 * The counts array must be obtained from
501 * vdev_indirect_mapping_load_obsolete_counts().
502 */
503 void
504 vdev_indirect_mapping_increment_obsolete_count(vdev_indirect_mapping_t *vim,
505 uint64_t offset, uint64_t length, uint32_t *counts)
506 {
507 vdev_indirect_mapping_entry_phys_t *mapping;
508 uint64_t index;
509
510 mapping = vdev_indirect_mapping_entry_for_offset(vim, offset);
511
512 ASSERT(length > 0);
513 ASSERT3P(mapping, !=, NULL);
514
515 index = mapping - vim->vim_entries;
516
517 while (length > 0) {
518 ASSERT3U(index, <, vdev_indirect_mapping_num_entries(vim));
519
520 uint64_t size = DVA_GET_ASIZE(&mapping->vimep_dst);
521 uint64_t inner_offset = offset -
522 DVA_MAPPING_GET_SRC_OFFSET(mapping);
523 VERIFY3U(inner_offset, <, size);
524 uint64_t inner_size = MIN(length, size - inner_offset);
525
526 VERIFY3U(counts[index] + inner_size, <=, size);
527 counts[index] += inner_size;
528
529 offset += inner_size;
530 length -= inner_size;
531 mapping++;
532 index++;
533 }
534 }
535
536 typedef struct load_obsolete_space_map_arg {
537 vdev_indirect_mapping_t *losma_vim;
538 uint32_t *losma_counts;
539 } load_obsolete_space_map_arg_t;
540
541 static int
542 load_obsolete_sm_callback(maptype_t type, uint64_t offset, uint64_t size,
543 void *arg)
544 {
545 load_obsolete_space_map_arg_t *losma = arg;
546 ASSERT3S(type, ==, SM_ALLOC);
547
548 vdev_indirect_mapping_increment_obsolete_count(losma->losma_vim,
549 offset, size, losma->losma_counts);
550
551 return (0);
552 }
553
554 /*
555 * Modify the counts (increment them) based on the spacemap.
556 */
557 void
558 vdev_indirect_mapping_load_obsolete_spacemap(vdev_indirect_mapping_t *vim,
559 uint32_t *counts, space_map_t *obsolete_space_sm)
560 {
561 load_obsolete_space_map_arg_t losma;
562 losma.losma_counts = counts;
563 losma.losma_vim = vim;
564 VERIFY0(space_map_iterate(obsolete_space_sm,
565 load_obsolete_sm_callback, &losma));
566 }
567
568 /*
569 * Read the obsolete counts from disk, returning them in an array.
570 */
571 uint32_t *
572 vdev_indirect_mapping_load_obsolete_counts(vdev_indirect_mapping_t *vim)
573 {
574 ASSERT(vdev_indirect_mapping_verify(vim));
575
576 uint64_t counts_size =
577 vim->vim_phys->vimp_num_entries * sizeof (uint32_t);
578 uint32_t *counts = vmem_alloc(counts_size, KM_SLEEP);
579 if (vim->vim_havecounts) {
580 VERIFY0(dmu_read(vim->vim_objset,
581 vim->vim_phys->vimp_counts_object,
582 0, counts_size,
583 counts, DMU_READ_PREFETCH));
584 } else {
585 bzero(counts, counts_size);
586 }
587 return (counts);
588 }
589
590 extern void
591 vdev_indirect_mapping_free_obsolete_counts(vdev_indirect_mapping_t *vim,
592 uint32_t *counts)
593 {
594 ASSERT(vdev_indirect_mapping_verify(vim));
595
596 vmem_free(counts, vim->vim_phys->vimp_num_entries * sizeof (uint32_t));
597 }
598
599 #if defined(_KERNEL)
600 EXPORT_SYMBOL(vdev_indirect_mapping_add_entries);
601 EXPORT_SYMBOL(vdev_indirect_mapping_alloc);
602 EXPORT_SYMBOL(vdev_indirect_mapping_bytes_mapped);
603 EXPORT_SYMBOL(vdev_indirect_mapping_close);
604 EXPORT_SYMBOL(vdev_indirect_mapping_entry_for_offset);
605 EXPORT_SYMBOL(vdev_indirect_mapping_entry_for_offset_or_next);
606 EXPORT_SYMBOL(vdev_indirect_mapping_free);
607 EXPORT_SYMBOL(vdev_indirect_mapping_free_obsolete_counts);
608 EXPORT_SYMBOL(vdev_indirect_mapping_increment_obsolete_count);
609 EXPORT_SYMBOL(vdev_indirect_mapping_load_obsolete_counts);
610 EXPORT_SYMBOL(vdev_indirect_mapping_load_obsolete_spacemap);
611 EXPORT_SYMBOL(vdev_indirect_mapping_max_offset);
612 EXPORT_SYMBOL(vdev_indirect_mapping_num_entries);
613 EXPORT_SYMBOL(vdev_indirect_mapping_object);
614 EXPORT_SYMBOL(vdev_indirect_mapping_open);
615 EXPORT_SYMBOL(vdev_indirect_mapping_size);
616 #endif